Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!ub!csn!news-1.csn.net!magnus.acs.ohio-state.edu!math.ohio-state.edu!uwm.edu!newsfeed.internetmci.com!in1.uu.net!bcstec!nntp
From: Mike Fouche <mikef@hsvaic.hv.boeing.com>
Subject: Re: Help: Improvement of Generalization of Feedforward ANN
X-Nntp-Posting-Host: e677744.hv.boeing.com
Content-Type: text/plain; charset=us-ascii
Message-ID: <DoM76C.5Kp@bcstec.ca.boeing.com>
Sender: nntp@bcstec.ca.boeing.com (NNTP News Access)
Content-Transfer-Encoding: 7bit
Organization: Missiles & Space Division
References: <4ih47a$gn5@wmwap1.math.uni-wuppertal.de> <4imj91$l8i@news.uni-c.dk>
Mime-Version: 1.0
Date: Thu, 21 Mar 1996 11:13:21 GMT
X-Mailer: Mozilla 1.2 (Windows; U; 16bit)
Lines: 16

Hello,

Hey I like that idea of weighted sum of the squared weights in the penalty function.
I've noticed for a long time that (for most cases) that neural weight sets where the
standard deviation was lower than the rest performed better in several areas.  
That's one of the things I look at when trying to decide which weight set to use
first.  I usually (a lot of "I"s in here ... bad writing :)) "eye" the weight sets
if I'm in a hurry and pick the one with the lowest deviation and smoothest 
weight/gain profile.

I'm going to implement this into our learning algorithm and see how it performs.  
You were helping Jens van Mahnen but you've helped me too!  Thanks!

Mike Fouche
Boeing Missiles & Space Division

